Update 3 New Gemini 3.6 Flash Models Make Other AI Look Outdated!
The map of the competition for artificial intelligence (AI) technology is heating up again. Without a magnificent celebration, the technology giant Google DeepMind has officially broken the market by announcing the three newest artificial intelligence models from the Gemini line. The three models are Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, as well as a special model of cybersecurity Gemini 3.5 Flash Cyber. Based on the official publication on Google Blog, this bold move marks the shift in Google’s focus from simply enlarging the size of the model parameters to improving cognitive efficiency, super high execution speed, and extreme computational cost cuts.
Adapted from the venturebeat technology media report, the Google maneuver was launched just ahead of the second quarter earnings announcement. The presence of the new Gemini trio directly targets the weak points of its competitors by offering high-level performance but with power consumption and a much more efficient number of tokens. As a result, AI agent-based automation can now run faster, smarter, and much more affordable for developers and large-scale corporations.
1. Gemini 3.6 Flash: An increasingly concise and smarter daily load horse
Based on the official Google Blog release, Gemini 3.6 Flash Positioned as a workhorse model or the main load horse to complete various complex daily tasks. Google designed this model directly from thousands of developer inputs using Gemini 3.5 Flash. The main input resolved in version 3.6 is the problem of token inefficiency and the AI response is often too long (verbose).
According to independent testing of Artificial Analysis Index, Gemini 3.6 Flash managed to reduce the use of output tokens by up to 17% compared to its predecessor. Even on long-term software engineering workflows such as benchmarks DataCurve Deepswe, the efficiency of the use of the token can be saved up to 65%. This means that AI can complete complex work with fewer thought cycles and much more direct execution on target.
Coding performance jump and multimodal reasoning
Citing the Google AI for Developers technical documentation, Gemini 3.6 Flash brings a number of significant architectural improvements:
- Ready-to-use code (production-ready code): Produces much more precise programming code, reduces bug fix cycles (debugging loops), and minimizes unnecessary file changes.
- Spatial reasoning & Visual: Sharp improvements in analyzing charts, reading visual blueprints (visual blueprints), to instantly converting images into multi-element web interface layouts.
- Computer Use Native Support: Equipped with the ability of the native to automate UI (User Interface), allowing AI agents to navigate and operate computer applications independently.
- Initial Programmatic Inspection: The model actively runs a diagnostic script first before changing the codebase, thereby increasing accuracy on large-scale programming tasks.
Adapted from the official benchmark data, Gemini’s 3.6 flash coding capability jumped by 49% on the test Deepswe (up from 37% on Gemini 3.5 flash). While in the research test of learning machines MLE-BENCH, the value skyrocketed to 63.9% (compared to 49.7% in the previous generation).
2. Gemini 3.5 Flash-Lite: The Fast Breaks the Computing Price Limit
If Gemini 3.6 Flash focuses on the balance of cognitive intelligence, then Gemini 3.5 Flash-Lite Present as the new speed king. based on measurement Artificial Analysis Index, this ultra-light model is capable of spewing responses of up to 350 token outputs per second!
Despite being labeled a lite, his cognitive abilities are surprising. Based on the release of Google DeepMind’s research, in several heavy AI agent benchmarks, Flash-Lite even managed to surpass the performance of the previous generation of high-end models such as Gemini 3 Flash. For example, in the coding test Swe-Bench Pro, Flash-Lite scored 54.2% (beaten Gemini 3 Flash which was at 49.6%). Likewise in the system navigation test Osworld-Verified, Flash-Lite won a landslide with a score of 74.0% compared to 65.1%.
Citing technical reviews from DataCamp, the combination of lightning speed and strong agent reasoning capabilities make the Gemini 3.5 Flash-Lite very suitable to be the engine of the automation sub-agent, mass document processing, and the AI mode feature that is starting to be rolled out in Google Search.
3. Gemini 3.5 Flash Cyber: Shield AI Special Fortress Security Code
The third family member that is no less interesting is Gemini 3.5 Flash Cyber. Adapted from the announcement on Google Blog, this model is specially designed to answer the increasingly complex challenges of modern cyber security.
Flash Cyber acts as the main brain behind CodeMender, Agent of code security automation made by Google. This particular model is trained to scan the code base massively, find vulnerability, verify vulnerabilities, to make code repairs (patching) automatically before the gap is exploited by irresponsible parties.
Based on Google’s official statement, in order to maintain security and prevent misuse of the cyber-offense feature, Gemini 3.5 Flash Cyber is not sold in general through public APIs. This model is provided exclusively to government agencies and trusted corporate partners through the CodeMender Pilot Program.
comparison of the latest gemini fire specifications and prices
Based on the financial report and analysis of TradingKey, Google’s aggressive pricing strategy is expected to put pressure on the margins of its competitors in the AI cloud industry. With the input context window capacity reaching 1 million tokens and an output limit of up to 64,000 tokens, here is a summary of the pricing scheme and its position:
| AI model name | input price (per 1m token) | Output price (per 1m token) | Speed/main features | Access availability |
|---|---|---|---|---|
| Gemini 3.6 Flash | $1.50 | $7.50 | very efficient token, precision coding, computer use | Fire Studio, Antigravity, Gemini App |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | Super fast (350 tokens/sec), mass automation | API Studio, Search AI Mode, Gemini App |
| Gemini 3.5 Flash Cyber | special enterprise | special enterprise | Patching Automatic Security Locks Via CodeMender | Exclusive Government & Proven Partners |
Developer API Update and Gemini Presence Signal 4
Citing official updates from Google AI for Developers Changelog, the launch of this new model also marks an overhaul of the Gemini API standard. Google has officially stopped using traditional sampling parameters such as temperature, TOP_P, and top_k. Instead, developers are directed to use configure the thinking levels like minimal or medium for more consistent results.
In addition, the entire new model has a knowledge base (knowledge cutoff) until March 2026. Interestingly, at the end of its official release, Google confirmed that they were conducting closed testing for the highest caste model, Gemini 3.5 Pro, while starting to design the initial foundation for future generation architecture, Gemini 4. For developers and users who want to try directly, Gemini 3.6 Flash and 3.5 Flash-Lite can already be accessed today via Google AI Studio, Android Studio, Google Antigravity, and the Gemini application on your smartphone!
